langfuse-mcp

Langfuse MCP is a Model Context Protocol server that connects Claude, Cursor, and other AI agents directly to Langfuse observability data. It provides 48 tools for inspecting traces, finding exceptions, analyzing sessions, managing prompts, and working with datasets without leaving your agent workflow. Operations teams use it to answer production questions: what failed, why was it slow, which prompt version ran, and what happened in a user's session. The server runs locally via stdio or HTTP and includes an agent skill with ready-made debugging playbooks. It focuses on compact trace inspection and token efficiency compared to broader hosted alternatives.

[{"step":"Set up the Langfuse MCP server and authenticate it with your Langfuse project. Use the server’s configuration to specify the project ID and API key.","action":"Run `langfuse-mcp-server --project-id your_project_id --api-key your_api_key` in your terminal or configure it in your agent’s environment."},{"step":"Identify the agent or trace you want to analyze. Use placeholders like [AGENT_NAME] or [TIME_PERIOD] in your prompt to scope the query.","action":"Replace [AGENT_NAME] with the exact name of your agent (e.g., 'CustomerSupportAgent') and [TIME_PERIOD] with a duration (e.g., 'last 3 days')."},{"step":"Define the metrics or thresholds to focus on. Common examples include latency, error rates, or tool call failures.","action":"Specify [SPECIFIC_METRIC] (e.g., 'latency') and [THRESHOLD] (e.g., '5 seconds') to filter traces automatically."},{"step":"Review the AI’s output and extract actionable insights. Cross-reference the trace data with your agent’s logs or code to validate findings.","action":"Use the suggested fixes to update your agent’s configuration, tooling, or prompts. Test the changes in a staging environment before deploying to production."},{"step":"Monitor the agent’s performance post-fix using the same Langfuse queries to confirm improvements.","action":"Re-run the query after 24-48 hours to compare latency, error rates, and user satisfaction metrics. Adjust thresholds or queries as needed."}]